Wireframe 3-D Solid

Physics & Simulation

A spinning polyhedron drawn live on the Super Nintendo — tetrahedron, cube, octahedron, icosahedron. Each frame builds a 3×3 rotation matrix (three Q8.8 matrix multiplies), projects every vertex through a perspective divide, and connects the edges with integer Bresenham lines into a 2bpp bitmap canvas. Written in C and compiled to a real .sfc ROM with the llvm-mos-based 65816 toolchain (+mos-a16 16-bit accumulator). Demo #16 of the compiler stress-test battery.

loading core…

↑/↓ spin Y · ←/→ spin X · Q/W dolly · X/A solid · S palette · Shift trail · Enter reset

Click the screen, then play. Switch solids with X/A; toggle trail mode with Shift.

Controls

Keys	Action
`↑ / ↓`	Speed Y-axis + / −
`← / →`	Speed X-axis + / −
`Q / W`	Dolly distance − / +
`A / X`	Previous / next solid (tetra → cube → octa → icosa)
`S`	Cycle palette
`Shift`	Toggle trail / crisp
`Enter`	Reset

The four solids

Cycle them with A (previous) / X (next). Each has its own vertex count and edge table — the icosahedron has 30 edges to draw per frame.

Solid	What it is
TETRA	4 faces, 4 vertices, 6 edges — the simplest solid
CUBE	6 faces, 8 vertices, 12 edges — a plain box in 3-D
OCTA	8 faces, 6 vertices, 12 edges — two square pyramids joined
ICOSA	20 faces, 12 vertices, 30 edges — the most complex

Rendering modes

Shift toggles between two modes shown in the top-right HUD:

TRAIL — phosphor accumulate: lines are never erased, leaving a luminous ghost. Try spinning fast then slow — the solid unpeels itself from its own trail.
CRISP — canvas clears each frame before redrawing: a clean, sharp rotation.

Tips

Set spin to zero with ↑/↓ and ←/→ to freeze the solid — then step it slowly.
Dolly in (Q) until the camera is almost inside the solid; the perspective distortion becomes extreme on the icosahedron.
It's a compiler stress-test. The mat3_mul loop calls __mulsi3 (32-bit software multiply) nine times per matrix element; the perspective divide calls __divsi3. Under +mos-a16 the 65816 runs in native 16-bit-accumulator mode, so the mode-switch rep/sep pairs bracket every multiply.

Written in C with an LLVM/65816 SNES toolchain. The 3-D math — matrix multiply + perspective divide — runs identically on the host oracle and all three compiler modes, verified against MAME and bsnes-jg. Hit Verify fidelity to reproduce the build gate's WRAM assert (3-D math hash 0xEF8E) live in this tab. No far pointers, so the same program is checked five ways: host == default == +mos-a16 == +mos-xy16.